github repo

What actually happened on Super Tuesday?

I’ve noticed that the press coverage of Super Tuesday of the 2020 Democratic primary seemed to focus almost exclusively on who “won” specific states. This is a familiar format of election coverage for most of us as its how we generally hear national results for general elections.

In the general election for president, almost every state is winner-take-all, meaning whoever receives a plurality of the vote in that state claims all of the electors from that state. Unlike the general election, in the Democratic primary, almost every state divides up its “delegates” sort of proportionally among all of the candidates who hit a certain threshold.

That means that presenting election results in terms of state wins can be misleading for a couple reasons: 1. It treats states where a single candidate gets all of the delegates, and states where delegates are equally split between two or more candidates exactly the same. 2. Because most of us don’t know the relative populations of all 50 states off the top of our heads, it makes it hard to compare candidate A’s strongest performing states with Candidate B’s.

The column on the left more accurately depicts where each candidate stands, while the chart on the right shows the stat underlying most election coverage.

Last night, you probably saw a map that looked similar to this on your preferred election coverage site.

When really, something like this better shows the intensity of victories in each state.

This doesn’t contradict the high-level narrative that Biden had a great night, but it paints a much more accurate picture of exactly how that great night went down.

Are these results surprising?

The big twist in the days leading up to Super Tuesday was two major candidates, Amy Klobuchar and Pete Buttigieg, both dropping out and endorsing Joe Biden. The results seemed to show that, perhaps unsurprisingly, Biden drastically outperformed the lead-up polling.

Assuming the polling wasn’t systematically wrong, this was probably due to a combination of Pete and Amy supporters voting for Biden because of their endorsements, and undecided voters choosing him as the standardbearer for the moderate wing of the party.

But what about Sanders? To the extent that only one candidate can get the nomination, Joe’s good night is Bernie’s bad night. But in terms of his actual share of the vote, Bernie met or outperformed his polling average (as calculated by 538) in every state.

That’s not to say that Sanders did well, just that he got about as many votes as the previous polling averages forecasted. The better question, then, is why did we expect him to do well?

This has at least a little to do with the overemphasis on “winning” states. Sanders was projected to win many Super Tuesday states, but was never really polling all that high in any of them, just high enough to keep his head above the rest of the pack. What seemed like a favorable position depended heavily on the moderate vote being split between several candidates. When candidates dropped out and their share mostly redistributed between those who remained, that lead shrunk.

It’s hard to say what the impact was of the coverage that inflated our expectations for Sanders. One the one hand, positive polling probably helped Sanders build enthusiasm and belief within his base. However, the abrupt shift away from that positivity may do more harm than good as he works to continue mobilizing supporters.

What might happen next?

If current polling averages are to be trusted (more on that later), Sanders’ path forward doesn’t look promising. Biden is polling ahead of him, both in terms of states and possible delegates, in most of the contests for which we have polling information.

The chart below compares polling averages for Biden and Sanders in upcoming contests. Note that this chart does not account for Klobuchar and Buttigieg supporters shifting to either candidate, as they overwhelmingly appear to have done for Biden on Tuesday. The actual reality, as things currently stand, is even bleaker for Sanders.

(Source: 538 Polling Averages. Excludes the following contests where polling was unavailable: LA, KY, PR, KS, NE, WV, RI, HI, ID, DC, SD, AK, ND, DA, GU, VI, MP.)

There is of course still another candidate in the race. Since Sanders looks pretty far out, lets look at one extreme scenario that some think could help him regain a footing.

The chart above entertains the idea of Warren dropping out and enthusiastically endorsing Sanders so effectively that all of her supporters switch to Sanders. Similar, it shows all Klobuchar and Buttigieg supporters in remaining states realigning to support Biden. This obviously won’t happen in totality, but it gives us some idea of how a potential endorsement would change things. This looks slightly less bleak for Sanders. A handful of states shift above the line, but the gap does grow larger in other large states (notably Florida).

To be clear, Warren hasn’t dropped out yet and hasn’t really indicated that she plans to. I won’t comment on how many of her supporters are likely to support Sanders even if she does drop out and endorse him, but to have a chance, Sanders would need it to be a lot.

As bleak as the picture is for Sanders, it’s worth noting that we should be less confident in these current estimates than we were for Super Tuesday states.

Should we take these projections seriously?

In a word, maybe? Even if we put aside the issue of potential realignment and just focus on the polling for Sanders and Biden, the answer isn’t really clear.

Unfortunately, 538 doesn’t seem to provide standard errors / margins of error / confidence intervals for their polling averages, but we can get some insight into their relative precision based on how often they’ve been polled.

On Super Tuesday, Sanders’ actual vote share was extremely close to his polling averages in most contests. However, not every state is polled the same amount and not every poll is conducted by the same quality of pollsters.

The early states were polled about as frequently as the upcoming contests relative to their delegate counts. However, 538 rates pollsters in terms of quality, and the upcoming contests have substantially fewer polls conducted by pollsters rated “B+” or better.

While almost two-thirds of all pledged delegates are still up for grabs, fewer than half of all polls conducted by quality pollsters were conducted on remaining states. This may cast some doubt on the poll averages of future contests, which seem to show the race is Biden’s to lose, but its difficult to quantify the extent.